Implementation of Virtual Try-On for Clothing Products Using Deep Learning Methods

Virtual Try-On technology has emerged as a break-through innovation in the e-commerce industry, allowing consumers to virtually try on clothing products before making a purchase. Deep learning methods, such as Generative Ad-versarial Networks (GANs), generate accurate simulations of how a garment will appear on the user's body. This technology addresses key challenges, such as uncertainties in clothing sizes and fits, which often hinder the online shopping experience. The development of virtual try-on continues to advance rapidly, enabling users to test products virtually before committing to a purchase. The outcomes of the virtual try-on model are evaluated based on achieved accuracy and visual results. Accu-racy is measured using Structural Similarity (SSIM) scores and Inception Score (IS). This study is based on previous research [1] that applied this technology to clothing using hyperparameters such as 20 epochs, a learning rate of 0.0002, and a batch size of 8 with a training dataset of 14,221 data. The previous study achieved an SSIM score of 0.845 and an IS score of 2.829. In this study, further development was conducted by testing different hyperparameters on smaller resources. The model with the highest accuracy was trained using hyperparameters of batch size 3, learning rate 0.0002, over 20 epochs, with a dataset of 1,000 training samples. This model achieved an SSIM score of 0.795071 and an IS score of 2.2008. The clothing items effectively covered the model's body during inference. The virtual try-on implementation was executed through a simple web interface using Flask.